independent speed test analysis of 4g mobile networks …€¦ ·...

Independent Speed test Analysis of 4G Mobile Networks Performed byDIKW Consulting

Hugo Koopmans

03-11-2015

1

Speedtest Analysis of 4G Mobile Networks

Contents

0.1 Colophon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

0.2 Code generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1 Abstract 5

2 Introduction 6

3 Step 1 : Data Collection 6

3.1 The raw data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.2 Speedtest.net data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.3 Sample test data from Ookla . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

4 Step 2: Data preprocessing 9

4.1 Basic data transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4.2 Suspicious devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4.3 Top three operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4.4 Focus on 4G technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4.5 Operating systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4.6 Geographical coverage area for 4G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

5 Step 3: Data analysis 18

5.1 Histogram distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5.2 Box-plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

6 Step 4: Test design 23

6.1 Which statistical test do we need? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

7 Step 5: Test results coverage area 25

8 Conclusion 26

9 Analysis and results Top 21 cities per city 27

9.1 Gemeente Amsterdam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

9.2 Gemeente Rotterdam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

9.3 Gemeente ’s-Gravenhage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

9.4 Gemeente Utrecht . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

9.5 Gemeente Eindhoven . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

9.6 Gemeente Tilburg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

9.7 Gemeente Groningen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

9.8 Gemeente Almere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

9.9 Gemeente Breda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

9.10 Gemeente Nijmegen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

The analysis was performed by DIKW Consulting on behalf of T-Mobile, based on Ookla’s Speedtest Intelligence Data

of 50


9.11 Gemeente Enschede . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

9.12 Gemeente Apeldoorn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

9.13 Gemeente Haarlem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

9.14 Gemeente Amersfoort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

9.15 Gemeente Arnhem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

9.16 Gemeente Zaanstad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

9.17 Gemeente Haarlemmermeer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

9.18 Gemeente ’s-Hertogenbosch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

9.19 Gemeente Zoetermeer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

9.20 Gemeente Zwolle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

9.21 Gemeente Maastricht . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

9.22 Gemeente Unkown Location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50


of 50


0.1 Colophon

This analysis is performed by DIKW Consulting.

DIKW Consulting is a consulting firm that takes her customers on the path from Data to information toKnowledge to Wisdom. Our expertise is in the field of data logistics, data warehousing, data mining andmachine learning.

T-Mobile has asked DIKW Consulting to perform this test as an independent third party. DIKWConsulting was paid to perform this test by T-mobile and has no other intentions then to perform thistest by it’s own high quality standards. The analysis was performed by generally accepted and approvedstandards and statistical methods using open source tools.

We let the data speak for itself.

If you have questions you can contact DIKW Consulting. If you want to repeat this test by yourself youare welcome to do so, all necessary scripts are available on GitHub. The data is commercially available atOokla.

This analysis, method, tools and scripts are open sourced and placed on GitHub, see the read-me on theGitHub repository.

For questions contact Hugo Koopmans at [email protected] or +31 6 43106780

0.2 Code generation

This is an R Markdown document. Markdown is a simple formatting syntax for authoring HTML, PDF,and MS Word documents. For more details on using R Markdown see http://rmarkdown.rstudio.com.


of 50

http://www.dikw.nl

http://www.ookla.com/

https://github.com/hugokoopmans/ookla-speedtest-analysis

mailto:[email protected]

http://rmarkdown.rstudio.com


1 Abstract

In this document we have conducted a statistical analysis of Ookla’s Speedtest Intelligence data fromspeedtest.net. Ookla provides commercially available speed test data collected by their mobile app on thethree main mobile platforms, Android, iOS and Windows mobile. We will load all the raw speed test datainto a database, analyse the data ot the top three operators(T-Mobile, Vodafone and KPN) and performa test on how fast their respective 4G networks are. Our test will be on download speed, upload speedand latency(or ping).

The speed test data that is provided by Ookla’s Speedtest Intelligence data from speedtest.net andobtained by T-Mobile is of July, August and September 2015. We will analyse and validate the datastep by step. After investigation on suspicious testing circumstances, such as, but not limited to, devices,location, theoretical maximum speeds and specific dates the data is ready to be subjected to a significancetest. This test will be done for all the data in the coverage area and for the biggest twenty cities in theNetherlands.

The result of these tests will give an answer to whether or not T-Mobile has on average a faster 4G networkthan Vodafone and/or KPN, based on the three metrics upload speed, download speed and latency.


of 50

http://www.speedtest.net/



2 Introduction

This document is a report of a statistical analysis of 4G network speed test data. The time period weconsider is Q3 2015 so the months July, August and September of 2015 are in scope. We perform ananalysis of Ookla Speedtest Intelligence data on three different measures:

• download speed• upload speed• latency(or ping)

On these metrics we compare the three major providers of 4G mobile networks in the Netherlands,Vodafone, KPN and T-Mobile.

This analysis is set up as follows:

• Step 1: Data Collection• Step 2: Data preprocessing

• Step 3: Data analysis• Step 4: Test design• Step 5: Test results coverage area

In section 8 we provide the conclusion, and additionally the analysis and results of the Top 20 cities percity is presented in section 9.

3 Step 1 : Data Collection

The data was downloaded from Ookla servers by T-Mobile. For this analysis we use a PostgreSQLdatabase that is locally installed.

The data is loaded from three different files, each file resembling a mobile platform (Android,iPhone andWindows mobile). The data is loaded as-is as it was received from the Ookla server. Scripts to load thisdata directly into a PostgreSQL database can be found in the GitHub repository.

All scripts to process the data are in SQL(available on GitHub) or R(included in this document, thanksto knitr)

Let’s get the data and do some basic counts.

3.1 The raw data

We load the raw files, downloaded from the Ookla server, into individual tables per mobile platform. Inthe table below we count the number of speed tests per mobile platform in Q3/2015.

Table 1: Raw test data counts

CountsAndroid 183501iOS 87184Windows 7688Sum 278373


of 50

http://www.ookla.com/

http://yihui.name/knitr/


So we start this analysis with in total 278373 speed tests, which are represented as rows in the data setcommercially downloaded from Ookla.

3.2 Speedtest.net data

Ookla designed their speed test in such a way that the results are as robust as possible. Ookla’sspeedtest.net is the de-facto standard for internet speed testing. According to http://www.ookla.com/speedtest-intelligence, Speedtest Intelligence is the choice of nearly every Fortune 500 ISP and MobileProvider in the world. For more information please visit Speedtest.net.

3.3 Sample test data from Ookla

Ookla has some random sample data available, this data can be used to validate our method. To validatethe test result one would need the specific data of the Netherlands.

A sample set of Ookla Speedtest Intelligence data can be found here. The files differ per mobile platform.The file descriptors for all three mobile platforms are listed below.

3.3.1 Android header descriptives

test_id - unique id of test in our systemdevice_id - unique device id in our systemandroid_fingerprinttest_date - YYYY-MM-DD HH:MM:SS in Pacific time (we can accommodate different time zones if needed)client_ip - ip of clientdownload_kbps - download speed in kilobits per secondupload_kbps - upload speed in kilobits per secondlatency_ms - ping in millisecondsserver_name - name of server tested to (name of city it is located in)server_country - country name of serverserver_country_code - country code of serverserver_latitude - latitude of server tested toserver_longitude - longitude of server tested toserver_sponsor_name - sponsor name of serverclient_country - country name of the clientclient_country_code - country code of the clientclient_region - region name of client (this will be state in the US)client_region_code - region code of clientclient_city - city of clientclient_latitude - latitude of client (GPS or Maxmind when location services disabled)client_longitude - longitude of client (GPS or Maxmind when location services disabled)miles_between - miles between the client and the server tested toconnection_type - http://developer.android.com/reference/android/telephony/TelephonyManager.html

0=unknown,1= Cell, 2=Wifi, 3=Gprs, 4=Edge, 5=Utms, 6=Cdma, 7=Evdo0, 8=EvdoA, 9=OnexRTT,10=Hsdpa, 11=Hspa, 12=Iden, 13=Ehrpd, 14=EvdoB, 15=Lte, 16=Hsupa, 17=Hspap

isp_name - name of ISP (Maxmind)is_isp - 0=Corporation/Academic, 1=ISPnetwork_operator_name - Mobile Carrier Name http://developer.android.com/reference/android

/telephony/TelephonyManager.html#getNetworkOperatorName()network_operator_code - MCC + MNC http://developer.android.com/reference/android

/telephony/TelephonyManager.html#getNetworkOperator()brand - http://developer.android.com/reference/android/os/Build.html#BRANDdevice - http://developer.android.com/reference/android/os/Build.html#DEVICEhardware - http://developer.android.com/reference/android/os/Build.html#HARDWAREbuild_id - http://developer.android.com/reference/android/os/Build.html#IDmanufacturer - http://developer.android.com/reference/android/os/Build.html#MANUFACTURERmodel - http://developer.android.com/reference/android/os/Build.html#MODELproduct - http://developer.android.com/reference/android/os/Build.html#PRODUCTcdma_cell_id - http://developer.android.com/reference/android/telephony/cdma/package-summary.htmlgsm_cell_id - http://developer.android.com/reference/android/telephony/gsm/package-summary.htmllocation_type - 0 = unknown, 1 = GPS, 2 = GeoIPsim_network_operator_name - Mobile Carrier Name from the SIMsim_network_operator_code - MCC + MNC from the SIM http://en.wikipedia.org/wiki/Mobile_Country_Code


of 50


http://www.ookla.com/speedtest-intelligence


https://support.speedtest.net/hc/en-us



3.3.2 iOS header descriptives

test_id - unique id of test in our systemdevice_id - unique device id in our systemtest_date - YYYY-MM-DD HH:MM:SS in Pacific time (we can accommodate different time zones if needed)client_ip - ip of clientdownload_kbps - download speed in kilobits per secondupload_kbps - upload speed in kilobits per secondlatency_ms - ping in millisecondsserver_name - name of server tested to (name of city it is located in)server_country - country name of serverserver_country_code - country code of serverserver_latitude - latitude of server tested toserver_longitude - longitude of server tested toserver_sponsor_name - sponsor name of serverclient_country - country name of the clientclient_country_code - country code of the clientclient_region - region name of client (this will be state in the US)client_region_code - region code of clientclient_city - city of clientclient_latitude - latitude of client (GPS or Maxmind when location services disabled)client_longitude - longitude of client (GPS or Maxmind when location services disabled)miles_between - miles between the client and the server tested toconnection_type - 0=unknown, 1=cell, 2=wifi, 3=GPRS, 4=Edge, 5=WCDMA, 6=HSDPA,

7=HSUPA, 8=CDMA1x, 9=CDMAEVDORev0, 10=CDMAEVDORevB, 11=eHRPD, 12=LTEisp_name - name of ISP (Maxmind)is_isp - 0=Corporation/Academic, 1=ISPcarrier_name - http://developer.apple.com/library/ios/documentation/NetworkingInternet

/Reference/CTCarrier/Reference/Reference.html#//apple_ref/occ/instp/CTCarrier/carrierNameiso_country_code - http://developer.apple.com/library/ios/documentation/NetworkingInternet

/Reference/CTCarrier/Reference/Reference.html#//apple_ref/occ/instp/CTCarrier/isoCountryCodemobile_country_code - http://developer.apple.com/library/ios/documentation/NetworkingInternet

/Reference/CTCarrier/Reference/Reference.html#//apple_ref/occ/instp/CTCarrier/mobileCountryCodemobile_network_code - http://developer.apple.com/library/ios/documentation/NetworkingInternet

/Reference/CTCarrier/Reference/Reference.html#//apple_ref/occ/instp/CTCarrier/mobileNetworkCodemodel - iPad, iPhone, iPod Touchversion - iOS versionlocation_type - 0 = unknown, 1 = GPS, 2 = GeoIP

3.3.3 Windows Mobile header descriptives

test_id - unique id of test in our systemdevice_id - unique device id in our systemtest_date - YYYY-MM-DD HH:MM:SS in Pacific time (we can accommodate different timezones if needed)client_ip - ip of clientdownload_kbps - download speed in kilobits per secondupload_kbps - upload speed in kilobits per secondlatency_ms - ping in millisecondsserver_name - name of server tested to (name of city it is located in)server_country - country name of serverserver_country_code - country code of serverserver_latitude - latitude of server tested toserver_longitude - longitude of server tested toserver_sponsor_name - sponsor name of serverclient_country - country name of the clientclient_country_code - country code of the clientclient_region - region name of client (this will be state in the US)client_region_code - region code of clientclient_city - city of clientclient_latitude - latitude of client (GPS or Maxmind when location services disabled)client_longitude - longitude of client (GPS or Maxmind when location services disabled)miles_between - miles between the client and the server tested toconnection_type - 0=unknown, 1=cell, 2=wifi, 3=GPRS, 4=1XRTT, 5=EVDO, 6=EDGE, 7=3G,

8=HSPA, 9=EVDV, 10=PassThru, 11=LTE, 12=EHRPDisp_name - name of ISP (Maxmind)is_isp - 0=Corporation/Academic, 1=ISPcarrier_name - AT&T, Verizon etcmanufacturer - Nokia, HTC, etc.device_name - name of the device for e.g. "HD7 T9292"hardware_version - device hardware version e.g. "1.0.0.0"firmware_version - device firmware_version e.g. "1232.2107.1241.1001"location_type - 0 = unknown, 1 = GPS, 2 = GeoIP


of 50


4 Step 2: Data preprocessing

In order to compare the data from the three different mobile platforms, we need to perform basic datatransformations and merge it into one table.

Following that, in this preprocessing and analysis step we validate the data on the following points: 1.Are there any specific individual devices that perform a suspiciously high number of tests? 2. We applyfilters so only the tests from the three operators we are interested in remain. 3. We apply filters so onlytests done on 4G technology remain. 4. We are only interested in the coverage area in which all threeoperators claim to have 4G coverage. 5. We look at speed test results that are “too good to be true”that is, measured speeds that are above the theoretical maximum possible for that specific technology.We remove these speed tests. 6. We look at specific coordinates that are very frequent, depending onthe explanation as for why these coordinates are used to often we remove or delete the speed tests percoordinate. 7. We look at specific dates that have a high number of speed tests for that day.

After all these checks we end up with a data set that is cleaned and ready to perform a statisticalsignificance test on the investigated metrics.

4.1 Basic data transformations

As explained in section 3.3., the data from the three different mobile platforms comes in different formatsand some basic data transformations are necessary before we can merge it into one table.

First of all, the names for the individual operators are spelled in various ways (e.g. ‘T-Mobile NL‘,‘T-Mobile NL’). Next, we need to map connection types to the specific technology used (2G, 3G or 4G)depending on the operating system of the device. For more details on these transformations, please seethe SQL script on GitHub.

After these transformations are performed, we proceed with the checks and cleaning steps explained at inthe previous section.

4.2 Suspicious devices

Are there any devices that perform tests very frequently?

In order to investigate this, let’s look at a frequency plot of devices that occur at least ten times in eachmonth. On the right hand side we see devices that are used for testing very often, some of them even onan hourly basis. Obviously those devices are not in the hands of real customers so these will be removedfrom the data set.


of 50


July 2015

Unique devices

Num

ber

of te

sts

010

020

030

040

0

August 2015

Unique devices

Num

ber

of te

sts

010

030

0

September 2015

Num

ber

of te

sts

020

040

0

Based on these plots and the amount of data consumed when performing a speedtest, we decide to removespecific devices that test more than 30 times a month. These devices are probably used by telecomprofessionals for testing purposes. The amount of data consumed per speedtest depends on the speed:the higher the speed, the more data is consumed. A test on 4G at very high speed can cost up to 100MBytes per test out of the data bundle. So, 30 or more tests per month at high speed are equivalent toapproximately 3GB of data usage just spend on tests alone and that is suspicious. The number 30 byitself is subjective, we could use 25 or 40 depending on what exact number of tests per device you wouldcall suspicious.

Actually including these are not influencing the results significantly, but we want to use real customerdata as much as possible, not affected by professionals testing their own (or others) network.

We identify 112 devices in July, 123 devices in August and 169 devices in September, which in totalrepresent 24138 speed tests. After filtering these devices the data set has 278373-24138=254235 speed testcases.

4.3 Top three operators

For this analysis we are only interested in the top three operators in the Netherlands. In the data set,at this point, there are 898 different operators identified. As we can see in the table below, in which weranked the ten most used operators, most of the speed tests were performed by people using one of thethree operators we are interested in. There are 7137 tests with no identifiable operator. We will filter outall but the top 3 operators and proceed with speed tests from these top three operators.

Table 2: Most frequent operators

Operator Number of speedtestsKPN NL 75152T-Mobile NL 51078Vodafone NL 47389DiGi 7423

7137TELKOMSEL 4640Tele2 NL 4167


of 50


Operator Number of speedtestsCarrier 3609TURKCELL 3538TIGO 2857

The top three operators together are good for 173619 tests conducted all over the Netherlands in the testperiod of Q3 2015. We keep only speed tests from the top three operators.

This leaves 254235 - 80616 = 173619 rows in the data set.

4.4 Focus on 4G technology

In the raw Ookla Speedtest Intelligence data the variable called ‘connection_type’ identifies whichtechnology is used, this variable can be transformed into the network technology used while performingthe test.

The variable Connection type defines 4G as connection type 15 for Android. For iOS connection type 12is LTE, and for Windows Mobile connection type 11 is LTE.

Definition of 4G for Android OS from the SQL script(available on GitHub) :

Case WHEN CONNECTION_TYPE=0 THEN 'UNKNOWN'WHEN CONNECTION_TYPE in (1,2) THEN 'WIFI/CELL'WHEN CONNECTION_TYPE in (3,4) THEN '2G'WHEN CONNECTION_TYPE=15 THEN '4G'WHEN CONNECTION_TYPE between 5 and 17 THEN '3G'ELSE 'UNKNOWN'

END AS TECHNOLOGY

Below we give an overview of the network technology types available in the data set.

Table 3: Technology used in tests

Number of cases Percentage2G 1906 1.103G 37344 21.514G 133927 77.14UNKNOWN 227 0.13WIFI/CELL 215 0.12

In the remainder of this analysis we will focus on 4G technology.

Filtering on 4G technology leaves 133927 test cases in the data set.

4.5 Operating systems

For the top three operators we can look at the type of operating system used on these devices:

KPN NL T-Mobile NL Vodafone NLAndroid 41896 25803 23426iOS 31048 24115 22497Windows 2208 1160 1466

Most of the tests were conducted on iOS closely followed by Android OS. Windows Mobile devices have


of 50


limited representation in the data set. In this test we are not interested in testing the difference inperformance per device or operating system.

4.6 Geographical coverage area for 4G

For this test to be fair to all three operators, we limit the comparison of the test to areas in which all threeoperators (KPN, Vodafone and T-Mobile) claim to have 4G coverage at the time of the measurements.While KPN and Vodafone already claim national 4G coverage, T-Mobile is still in the process of expandingtheir 4G network. Therefore, T-Mobile 4G coverage area is extended every month. This means thatsome areas only got 4G coverage during Q3 2015, the period of the test. All of the top 21 cities in theNetherlands, which we will analyze per city, had 4G enabled prior to Q3 2015.

4.6.1 Coordinates with very high number of tests

Are there any locations, or coordinates, that occur very often in the investigated 4G area? If we join thecoordinates latitude and longitude together and look at the most frequent occurrences we see that thereare indeed some coordinates that are very frequent. How do these exact same coordinates end up in thedata? To understand this we need to explain a bit more on how the Ookla Speed test application gets thecoordinates from a mobile device(read more online). There are several scenario’s that can be the case:1)The customer has approved the application access to the GPS coordinates of his/her device. 2)For somereason the app cannot read the GPS coordinates from the device at the time of the test. This reason canbe of different origins, the user has blocked access or we are in a building or there are other technicalreasons why the exact GPS coordinates cannot be accessed.

Whenever the exact coordinates are not available, due to measurement issues or because the customer isnot allowing the application to use the GPS coordinates Ookla uses GEO-IP. GEO-IP is a online serviceto estimate the physical location of an ip-internet address (more online from maxmind ).

Coordinates Count52.3667 , 4.9 3160552.374 , 4.8897 164052.3666 , 4.9027 82651.9225 , 4.4792 17052.0767 , 4.2986 12252.35 , 4.9167 11852.1583 , 4.4931 8051.4408 , 5.4778 7152 , 6 6752.3808 , 4.6368 6551.8425 , 5.8528 6452.056 , 4.4983 6351.81 , 4.6736 6152.0833 , 5.0833 5951.9167 , 4.4333 57

We asked for an explanation from Ookla on frequent and rounded coordinates.Coordinate (52.3667,4.9) : Response from Ookla: Results are definitely coming from GEO-IP Coordinatesas you know. 2. In a single day here, there is 191 results from the same IP block to this location. 3.Correct, this is similar to the ‘Kansas’ issue we discussed last year.

Coordinate (52.35-4.9167) is in Amsterdam. Question to Ookla: Can we still assume the measurementsare from Amsterdam? Response Ookla: Yes, GeoIP is used here. We can assume you are in Amsterdambut like any other GeoIP location result, the confidence level isn’t as high as it would be if we were ableto get location information directly from the device, meaning if we were able to obtain GPS instead ofGEO-IP.

Coordinate (52.3666-4.9027) is also in Amsterdam(Waterloo plein).


of 50

https://support.speedtest.net/hc/en-us/articles/203845480-Mobile-Test-Server-Selection

https://www.maxmind.com/en/geoip2-services-and-databases


Coordinate (52.374-4.8897) is also in Amsterdam(Spuistraat).

Coordinate (51.9167-4.5) is in Rotterdam. Again with limited precision, same response as above fromOokla.

Coordinate (52.0666-4.3209) is in The Hague. All cases are tests with operator equal to “T-Mobile NL”also this location is close to the T-Mobile office in The Hague. We will exclude these tests as potentiallybeing from T-Mobile employees.

Also see the precision of the coordinates denotes the fact that we are unsure about the exact location.

So what do we do with these suspicious coordinates? The “Unknown” location, which hascoordinates (52.3667 , 4.9) and is the default location from Ookla if the GPS cannot pick up the exactlocation during the test, we have 31605 speed tests with this coordinate alone. We will exclude testsperformed at this coordinate from the general analysis. Nevertheless, we will analyse this set the sameway we analyse individual cities, the result for this set can be found as the last city labeled “unknown”.

Head office locations We removed 47 tests from coordinates “52.0666 , 4.3209” in The Hague. As theyare close to the head-office of T-Mobile and indeed all tests from this location are done from a T-Mobilenetwork.

We did the same for locations close to the head offices of KPN and Vodafone. We removed 14 tests fromthe co-ordinates “52.0666 4.3479” (KPN office) and we removed 5 tests from the coordinates “52.37674.9061”" (Vodafone office Amsterdam).

After removing 66 tests from the coordinates around the head-offices of the top three providers as describedin the above paragraph, together with 31605 from the coordinates “52.3667 , 4.9” the data set contains133927 - 31671 = 102256 speed tests at this point.

4.6.2 Mapping test coordiantes to city boundaries

To identify the exact 4G coverage area we will use in this analysis, we use data from CBS. CBS is theCentral Bureau of Statistics in the Netherlands. They provide publicly available polygon data on cities inthe Netherlands. Based on these geographical city boundaries we map each latitude, longitude coordinateonto a city.

From T-Mobile we received a list of cities that, at the time of testing, have 4G coverage. From the datawe can see that per city each provider has sufficient number of speed tests in the data set for the tests tobe representative.

We use city boundaries to not be influenced by exact locations of network infrastructure.

Online we can get an up to date overview of network coverage for all three operators (see here).

This test is not about coverage but about speed of the 4G network.

T-Mobile has the least 4G coverage of the three operators and per end of Q3 2015 has actual coverage inthe following area:


of 50

https://en.wikipedia.org/wiki/Decimal_degrees

http://www.cbs.nl/nl-NL/menu/themas/dossiers/nederland-regionaal/publicaties/geografische-data/archief/2014/2013-wijk-en-buurtkaart-art.htm

http://www.4gdekking.nl/


In order to do a fair comparison, we focus only on the area presented in the picture above.

We do a Geo-spacial filter on the latitude and longitude coordinates provided in the data set and onlykeep speed tests that are in any of the above city boundaries. For the area, defined above, the followingnumber of tests are available in the data set.

Table 6: Number of 4G speedtests in the selected coverage area.

Number of tests percentageKPN NL 39599 40.74T-Mobile NL 31740 32.66Vodafone NL 25856 26.60Sum 97195 100.00

Areas where T-Mobile announced 4G coverage in Q3 2015:

July: Heerenveen, Weststellingwerf, Hoogeveen, Meppel, Staphorst, Asten, Boekel, Deurne, Someren,Venray, Gemert-Bakel, Heeze-Leende, De Wolden, Westerveld and Midden-Drenthe

August: Leeuwarden, Graft-De Rijp, Beemster, Bergen (NH.), Castricum, Edam-Volendam, Enkhuizen,Heerhugowaard, Heiloo, Den Helder, Hoorn, Langedijk, Medemblik, Opmeer, Schagen, Texel, Schermer,Zeevang, Drechterland, Stede Broec, Waterland, Beesel, Nederweert, Venlo, Weert, Horst aan de Maas,Koggenland, Leudal, Cranendonck, Peel en Maas and Hollands Kroon

September: Emmen, Zeewolde, Skarsterlân, Lemsterland, Kampen, Noordoostpolder, Urk, Elburg, Olde-broek, Dronten, Baarle-Nassau, Etten-Leur, Rucphen, Steenbergen, Woensdrecht, Zundert, Halderberge,Roosendaal, Steenwijkerland, Alphen-Chaam and Zwartewaterland

All other cities in the TMNL coverage area had 4G enabled prior to Q3 2015.

Naturally, in the cities where 4G was announced in Q3, almost all of T-Mobile’s 4G speedtests occur afterannouncing to customers that 4G is activated, so July, August or September respectively. Even though avery small number of tests were executed during the extension of the 4G coverage onto these cities, as4G sites were being added (so before announcing 4G coverage in these cities; this is possible as for most4G capable phones, the 4G network is selected automatically if available). In the same cities, KPN andVodafone speedtests are distributed more evenly over the entire period of Q3/2015. However, this has no


of 50


influence on the test results, as the test results per month do not differ substantially (please see section4.6.4).

So we filtered the data set to include only speed tests from the coverage area, speed tests outside this areaare neglected. The data set now contains 102256 - 5061 = 97195 speed tests.

4.6.3 Suspicious speeds

In the data we check for up and download speeds that are technically impossible. Download speeds for 4Gare limited to 150Mbps on the T-mobile technology.

KPN and Vodafone have a technology called LTE advanced which has a maximum download speed of225Mbps.

Any speed tests that had a speed recorded above the technical maximum for that operator was removedfrom the data set.

So we remove suspicious measurements in which the download speed exceeded the maximum theoreticalspeed per individual operator.

For T-Mobile we removed 11 cases, for Vodafone we removed 3 cases and for KPN we removed 7 cases,because they where above 150(or 225) MBps.

After removing in total 21 of these suspicious measurements, the data set contains 97174 speed tests atthis point.

Let’s do the same for upload speed.

The maximum theoretical upload speed is for all operators the same: maximum upload speed of 50Mbps.

So again we remove suspicious measurements in which the upload speed exceeded the maximum theoreticalspeed per individual operator. For T-Mobile we removed 35 cases, for Vodafone we removed 33 cases andfor KPN we removed 62 cases.

After removing in total 130 of these suspicious measurements the data set contains 97044 speed tests atthis point.

Let’s look at latency also. We see the maximum value in latecy measurements is 65535, which is themaximum value for a variable of type unsigned short. This is clearly an outlier and should be removed.

Response Ookla: “That issue was resolved in the latest Android release. As more and more clientsdownload the latest version of the app, this will become less and less obvious. Just filter out for now untilits cleaned up in the files.”

So we remove 4 measurements from the T-Mobile set , 5 measurements from the KPN set and 0measurements from the Vodafone set. After this the data set contains 97035 speed tests at this point.


of 50


4.6.4 Suspicious dates or times

We count the number of tests per day for the months July, August and September of 2015. Again we arelooking at any suspicious peaks in the data.

KPN NL

T−Mobile NL

Vodafone NL

0200400600800

0200400600800

0200400600800

2015

−07

−01

2015

−07

−02

2015

−07

−03

2015

−07

−04

2015

−07

−05

2015

−07

−06

2015

−07

−07

2015

−07

−08

2015

−07

−09

2015

−07

−10

2015

−07

−11

2015

−07

−12

2015

−07

−13

2015

−07

−14

2015

−07

−15

2015

−07

−16

2015

−07

−17

2015

−07

−18

2015

−07

−19

2015

−07

−20

2015

−07

−21

2015

−07

−22

2015

−07

−23

2015

−07

−24

2015

−07

−25

2015

−07

−26

2015

−07

−27

2015

−07

−28

2015

−07

−29

2015

−07

−30

2015

−07

−31

2015

−08

−01

2015

−08

−02

2015

−08

−03

2015

−08

−04

2015

−08

−05

2015

−08

−06

2015

−08

−07

2015

−08

−08

2015

−08

−09

2015

−08

−10

2015

−08

−11

2015

−08

−12

2015

−08

−13

2015

−08

−14

2015

−08

−15

2015

−08

−16

2015

−08

−17

2015

−08

−18

2015

−08

−19

2015

−08

−20

2015

−08

−21

2015

−08

−22

2015

−08

−23

2015

−08

−24

2015

−08

−25

2015

−08

−26

2015

−08

−27

2015

−08

−28

2015

−08

−29

2015

−08

−30

2015

−08

−31

2015

−09

−01

2015

−09

−02

2015

−09

−03

2015

−09

−04

2015

−09

−05

2015

−09

−06

2015

−09

−07

2015

−09

−08

2015

−09

−09

2015

−09

−10

2015

−09

−11

2015

−09

−12

2015

−09

−13

2015

−09

−14

2015

−09

−15

2015

−09

−16

2015

−09

−17

2015

−09

−18

2015

−09

−19

2015

−09

−20

2015

−09

−21

2015

−09

−22

2015

−09

−23

2015

−09

−24

2015

−09

−25

2015

−09

−26

2015

−09

−27

2015

−09

−28

2015

−09

−29

2015

−09

−30

date

freq

We find no disturbing or unknown peaks on a specific date, except for Vodafone on the 18/08/2015. Thiscan be explained by the network incident that Vodafone had on this date, causing customers to experienceslow or no internet. The problem was fixed the same day and it was communicated to customers, which isa stimulus to test.

Table 7: Counts per operator per month

07 juli 08 augustus 09 september SumKPN NL 12663 12907 13955 39525T-Mobile NL 9774 11175 10741 31690Vodafone NL 7895 9206 8719 25820Sum 30332 33288 33415 97035

Above a count per month per operator in the 4G coverage area. This is the data set on which we conductthe remainder of the analysis.

In the tables below we list the averages for download speed, upload speed and latency per operator permonth. For KPN we see a large increase in download and upload speed between August and September,as well as reduced latency. This is probably a result of KPN adding a substantial amount of sites in thelarge cities by activating 4G on the 1800Mhz frequency. This also enables them to activate LTE Advancedon those locations, which has an additional positive influence on average download speed, because it allowsdownload speeds of up to 225Mbps, provided that the customer doing the tests has a device that cansupport this and the network conditions at the location allow this.

Table 8: Average download speed(Kbps) per operator per month

07 juli 08 augustus 09 septemberKPN NL 32171.9 32089.7 35748.1T-Mobile NL 42315.8 42262.6 41078.1Vodafone NL 29682.1 28933.5 29970.1


of 50


Table below shows details for upload speed.

Table 9: Average upload speed(Kbps) per operator per month


Table below shows details for latency.

Table 10: Average latency(ms) per operator per month



of 50


5 Step 3: Data analysis

So we have pre-processed the data and looked for anomalies in the data. If found, we have correctedthem. Finally we are ready to compare speed test data between the three major telecom operators inNetherlands in the above defined coverage area for the period of Q3 2015.

We will analyse three different metrics:

• Download speed• Upload speed• Latency

There is no useful way to aggregate these individual metrics into one overall ‘speed aggregated score’.Most customers are interested in download speeds, because it affects the most of their experience(browsing,streaming,downloading etc). Most network speed comparisons only focus on download speed.However, since upload speed is also important for posting video’s and photos on social media, and pingtimes are important for gaming and fast opening of websites these metrics are also analyzed.

So how are the different metrics distributed?

5.1 Histogram distributions

A histogram is a graphical representation of the distribution of data. It is an estimate of the probabilitydistribution of a continuous variable (quantitative variable), more about histograms on Wikipedia.

5.1.1 Download speed

In the histogram below we see download speed in Kbps on the horizontal axis. The number of test casesare plotted as bars, on the vertical axis we see the count of the number of speed tests in a specific bin (orrange). The histogram gives a visual representation of the distribution of the data.

0

2000

4000

0

2000

4000

0

2000

4000

KP

N N

LT

−M

obile NLV

odafone NL

0 50000 100000 150000 200000Download speed (Kbps)

Num

ber

of s

peed

test

s (c

ount

)

We plot the three histograms, one for each operator, right above each other so the horizontal axes arealigned. The red lines denotes the mean (or average) speed of all speed tests for the specific operator.If the red line is placed to the righthandside in this histogram that means the average speed for thisoperator is faster. The red line to the left means the avarage speed is slower. We can see that the red lineof T-Mobile is farthest to the right so T-Mobile apparently has the highest average download speed.


of 50

https://en.wikipedia.org/wiki/Histogram


5.1.2 Upload speed

In the histogram below we see upload speed in Kbps on the horizontal axis. The number of test cases areplotted as bars, on the vertical axis we see the count of the number of speed tests in a specific bin (orrange). The histogram gives a visual representation of the distribution of the data.

0500

100015002000

0500

100015002000

0500

100015002000

KP

N N

LT

−M

obile NLV

odafone NL

0 10000 20000 30000 40000 50000Upload speed (Kbps)

Num

ber

of s

peed

test

s (c

ount

)

We plot the three histograms, one for each operator, right above each other so the horizontal axes arealigned. The red lines denotes the mean (or average) speed of all speed tests for the specific operator.If the red line is placed to the righthandside in this histogram that means the average speed for thisoperator is faster. The red line to the left means the avarage speed is slower. We can see that the red lineof T-Mobile is farthest to the right so T-Mobile apparently has the highest average upload speed as well.

5.1.3 Latency

In the histogram below we see latency speed on the horizontal axis on a logaritmic scale. The logtransformation makes the figure more readable. The number of test cases are plotted as bars, on thevertical axis we see the count of the number of speed tests in a specific range. For latency we take the logso the outlines scale and we can have a look at the shape of the distribution.

010002000300040005000

010002000300040005000

010002000300040005000

KP

N N

LT

−M

obile NLV

odafone NL

10 100 1000Latency (ms) log scale

Num

ber

of s

peed

test

s (c

ount

)

For latency smaller is better, so in this plot we are looking at which operator has the smallest latency.Again we see the red lines per operator. The average latency(red line) for T-Mobile is the most to the left,which means T-Mobile has the smallest average latency of the three operators.


of 50


5.2 Box-plot

In descriptive statistics, a box plot or box-plot is a convenient way of graphically depicting groups ofnumerical data through their quartiles. Box plots may also have lines extending vertically from the boxes(whiskers) indicating variability outside the upper and lower quartiles, hence the terms box-and-whiskerplot and box-and-whisker diagram. Outliers may be plotted as individual points. More about box-plotson Wikipedia.

5.2.1 Download speed

We see a box plot for each of the operators on the x axis, the vertical axis shows the speed test values inKbps. The daimond shape represents the mean, the thick line represents the median. the black dots onthe top and bottom represent extreme cases. The data is split up into quartiles, that means four equallysized proportions. The first and fourth quarter are represented as a line, the second and third as a box.

0

50000

100000

150000

200000

KPN NL T−Mobile NL Vodafone NLoperator

Dow

nloa

d sp

eed

(Kbp

s)

Also in the box-plot we see that for T-Mobile the diamond shape(average) and the thick line representingthe median are higher then the same values for the other operators. So also in this box-plot for downloadspeed we see that T-Mobile has the highest average and median download speed.

If we zoom in on percentiles we can look at the fastest 10%, 5% and 1% of speed tests:

Table 11: Top percentiles average download speed(Kbps)

operator 10% 5% 1%KPN NL 69698 92459 135121T-Mobile NL 81151 92810 113986Vodafone NL 62579 85027 117379

From the table we see that T-Mobile scores best in the 10% of download speed tests per operator. For the5% percentile the T-Mobile outcome is very close to the KPN outcome but both better then Vodafone.

In the top 1% of download speed tests KPN scores best, followed by Vodafone. Both these operatorsthey have the LTE-Advanced technology in some locations mostly in the largest cities. This technology issupported by a small number of devices. In excellent radio conditions and in case the user has a suitabledevice, download speeds higher than 150MBps (up to 225Mbps) can be achieved. This is the case inless than 1% of all tests of KPN and Vodafone, but also the main reason why the top 1% for these twooperators is higher than T-Mobile’s. The LTE-advanced technology does not contribute to higher uploadspeeds nor lower latency.


of 50

https://en.wikipedia.org/wiki/Box_plot


5.2.2 Upload speed


0

10000

20000

30000

40000

50000


Upl

oad

spee

d (K

bps)

Also in the box-plot we see that for T-Mobile the diamond shape(average) and the thick line representingthe median are higher then the same values for the other operators. So also in this box-plot for uploadspeed we see that T-Mobile has the highest average and median upload speed.

If we zoom in on percentiles we can look at the fastest 10%, 5% and 1% of speed tests:

Table 12: Top percentiles average upload speed(Kbps)


From the table we see that T-Mobile scores best in the fastest 10%,5% and 1% of upload speed tests peroperator.

5.2.3 Latency

For latency ( or ping) we take the log so the outliers scale and we can have a look at the shape of thedistribution.



of 50


10

100

1000


Late

ncy

(ms)

log

scal

e

In the box-plot for latancy we see that for T-Mobile the diamond shape(average) and the thick linerepresenting the median are lower then the same values for the other operators. Remember, for latancylower is better.

If we look at the fastest 1%, 5% and 10% percentiles we see the following table:

Table 13: Top percentiles average latency (ms)



of 50


6 Step 4: Test design

What we want to test is if, on average, a customer that uses T-Mobile 4G Mobile Network gets a higheraverage speed(in terms of download speed, upload speed and latency) than a customer using KPN’s orVodafone’s 4G Mobile Network, with all else equal. To do this we have collected thousands of OoklaSpeedtest results taken from the three top operators (KPN, Vodafone and T-Mobile) which have beenfiltered as set out in the above to a final dataset consisting of 97035 data points. Now we want to compareT-Mobile with the other two operators. So we do two tests: the first is comparing T-Mobile with KPNand the second is comparing T-Mobile with Vodafone. In each test we compare all three metrics: Uploadspeed, Download speed and latency(or ping). For each operator we have a sample set available in thedata. These sets are so called samples (from the Dutch population of mobile phone users) from which wecalculate the sample means. Now our statistical test tests if these sample means are significantly differentfrom one another.

In practice, the Central Limit Theorem assures us that, under a wide range of assumptions, the distributionsof the two sample means being tested will themselves approach Normal distributions as the sample sizesget large, regardless (this is where the assumptions come in) of the distributions of the underlying data.As a consequence, as the sample size gets larger, the difference of the means becomes normally distributed,and the requirements necessary for the t-statistic of an unpaired t-test to have the nominal t distributionbecome satisfied.

6.1 Which statistical test do we need?

What we have here is a set of unpaired, independent, different sample size, different variance data. Asuitable and powerful test for this kind of data is a Welch t-test.

In statistics, Welch’s t-test (or Welch-Aspin Test) is a two-sample location test, and is used to check thehypothesis that two populations have equal means(our NULL hypothesis). Welch’s t-test is an adaptationof Student’s t-test, and is intended for use when the two samples have possibly unequal variances(whichis the case here). These tests are often referred to as “unpaired” or “independent samples” t-tests, asthey are typically applied when the statistical units underlying the two samples being compared arenon-overlapping(in our case the units are different people performing the test with different devices ondifferent networks).

6.1.1 Significance

So when is a test significant? And if so at what level? And furthermore can we qualify such a significantresult as good or bad? To start with the last remark, all qualifications of a statistical result are subjective.One way of looking at 95% confidence is that 1 out of 20 trials (in 5% of the cases) you make a so calledType 1 error, in which you wrongly reject the null-hypothesis. So in this case, if the p-value would be0.05(confidence level 95%) you would claim that operator x is faster then operator y while in fact theywere not. In applied practice, confidence intervals are typically stated at the 95% or 99% confidence level(More on significance ).

In our test we will set the confidence level to be 99%, which is more strict then 95%. This means we willreject the Null Hypothesis only if we are 99 % confident we do not make a mistake. From the test resultwe see that in most cases the calculated p-values are very much smaller then 1 - 0.99 = 0.01, so changesof making this type of error are even considerably smaller than the claimed confidence level of 99%.

6.1.2 P-value

In statistics, the p-value is a function of the observed sample results (a statistic) that is used for testinga statistical hypothesis. Before performing the test a threshold value is chosen, called the significancelevel of the test, traditionally 5% or 1% and denoted as α. If the p-value is equal or smaller than thesignificance level (α), it suggests that the observed data is inconsistent with the assumption that the nullhypothesis is true, and thus that hypothesis must be rejected and the alternative hypothesis is acceptedas true ( see wikipedia).


of 50

https://en.wikipedia.org/wiki/Welch%27s_t_test

https://en.wikipedia.org/wiki/Statistical_significance

https://en.wikipedia.org/wiki/P-value


6.1.3 Confidence intervals

Confidence intervals consist of a range of values (interval) that act as good estimates of the unknownpopulation parameter. The level of confidence of the confidence interval would indicate the probabilitythat the confidence range captures this true population parameter given a distribution of samples. Thisvalue is represented by a percentage, so when we say, “we are 99% confident that the true value of theparameter is in our confidence interval”, we express that 99% of the observed confidence intervals willhold the true value of the parameter. A confidence interval does not predict that the true value of theparameter has a particular probability of being in the confidence interval given the data actually obtained.(see wikipedia).


of 50

https://en.wikipedia.org/wiki/Confidence_interval


7 Step 5: Test results coverage area

We test T-Mobile against the other two operators so we have two tests. We put the confidence level to 99%. Our null-hypothesis is that the means are drawn from the same sample, so they are not different.

In this test we use the whole cleaned data set, in the next chapter we test each individual city.

Let’s see what our test results are:

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 31690 41878 39525 33408 < 2.22e-16 Yes 8469.8 +/- 529 25.4T-Mobile Vodafone 31690 41878 25820 29512 < 2.22e-16 Yes 12365.1 +/- 556.2 41.9

Table 14: Comparison of means for metric: Download (Kbps)

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 31690 17501 39525 14198 < 2.22e-16 Yes 3303 +/- 236.5 23.3T-Mobile Vodafone 31690 17501 25820 14045 < 2.22e-16 Yes 3455.9 +/- 263.2 24.6

Table 15: Comparison of means for metric: Upload (Kbps)

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 31690 36.6 39525 41.8 < 2.22e-16 Yes -5.2 +/- 1.1 -12.4T-Mobile Vodafone 31690 36.6 25820 45.6 < 2.22e-16 Yes -9 +/- 1.2 -19.7

Table 16: Comparison of means for metric: Latency (ms)

Explanation of termsSample 1: Number of speed test samples for operator 1.

Sample 2: Number of speed test samples for operator 2.

Mean 1: Average speed of speed tests for operator 1 in Kbps. A high number here means that this operator has a fastdownload(or upload) speed. For the latency a high number means you have to wait longer to contact webpages or servers.

Mean 2: Average speed of speed tests for operator 2 in Kbps. A high number here means that this operator has a fastdownload(or upload) speed. For the latency a high number means you have to wait longer to contact webpages or servers.

P-value: The test statistic, for more explanation see the paragraph P-Value above

Sign.: Short for Significance. We compare the test statistic with the predefined confidence level(0.99). ‘Yes’ means the testis significant, ‘No’ means the test is not significant.“))

Diff(in Kbps or ms): Difference of the means (DoM) is the difference of Mean 1 and Mean 2(Mean 1 - Mean 2). Fordownload and upload speeds(Kbps) big positive number here means operator 1 has a faster speed then operator 2. A bignegative number means that operator 2 has a faster speed then operator 1. For latency(ms) this is the opposite, becausesmaller is better.

Conf Int: Confidence interval consist of a range of values (interval) that act as good estimate of the true difference of themean. We are 99% confident that the true value of the difference of the mean is in our confidence interval.

Rel(%): Relative difference in percentage. It is calculated as the difference of the mean divided by the slower of the twooperators average speed. If the difference is not significant(column Sign is No), this column will state NA(Not Applicable).The comparison rules are similar to what is explained in the Diff(in Kbps or ms).

Looking at the tables above we see that all results are significant at α = 0.01 level(99% confidencelevel) and the resulting p-values are very small. This means we can reject the null-hypothesis with greatconfidence. Hence we can state that with 99% confidence the true difference in the means lies within theconfidence interval provided in the table.


of 50


8 Conclusion

This analysis has been conducted with the utmost care and to the best knowledge of the analyst (DIKWConsulting). The analysis is opensource and all code can be downloaded, reviewed and repeated fromGitHub.

Overall we can say that based on the speedtest data analysed in the investigated area the 4G network ofT-Mobile outperforms both KPN and Vodafone on download speed, upload speed and latency.

From the data analysed in the investigated area the average download speed of the 4G network of T-Mobileoutperforms KPN by 8.47 Mbps, which is 25.4%. Also, from the data analysed in the investigated areathe average download speed of the 4G network of T-Mobile outperforms Vodafone by 12.37 Mbps, whichis 41.9%. From table 14 above similar statements can be derived for upload speed. For deriving thesestatements for latency, please see table 15 keeping in mind that smaller values are better.

For conclusions per individual city we refer to the section below. Please keep in mind that the significanceof a test per city does not influence the significance of a test over the whole 4G area. The significance of atest per city only shows if the 4G network speeds (download speed, upload speed and latency) are alsosignificantly different on a local level, so for that city treated separately.


of 50

https://github.com/hugokoopmans/ookla-speedtest-analysis


9 Analysis and results Top 21 cities per city

From CBS we have the following top 21 cities based on number of inhabitants(“aantal inwoners”) from2014. We see that Zwolle is now number twenty in favour of Maastricht(now 21). For completness wekeep Maastricht in the list of cities to be analysed.

code naam aantal inwoners oppervlak(km2)GM0363 Amsterdam 810935 165.89GM0599 Rotterdam 618355 208.88GM0518 ’s-Gravenhage 508940 81.87GM0344 Utrecht 328165 94.26GM0772 Eindhoven 220920 87.67GM0855 Tilburg 210270 117.24GM0014 Groningen 198315 78.27GM0034 Almere 196010 129.30GM0758 Breda 179620 126.05GM0268 Nijmegen 168290 53.61GM0153 Enschede 158585 140.99GM0200 Apeldoorn 157545 339.88GM0392 Haarlem 155145 29.22GM0307 Amersfoort 150895 62.82GM0202 Arnhem 150820 97.96GM0479 Zaanstad 150595 73.92GM0394 Haarlemmermeer 144060 178.68GM0796 ’s-Hertogenbosch 143730 84.40GM0637 Zoetermeer 123560 34.53GM0193 Zwolle 123160 111.31GM0935 Maastricht 122485 56.62

In this test we use the CBS cities in this area as the benchmark area for the overall comparison.

In more detailed analysis we investigate the top twenty “Gemeentes” based on number of inhabitants.based on data available at CBS.


of 50

http://www.cbs.nl/nl-NL/menu/themas/dossiers/nederland-regionaal/publicaties/geografische-data/archief/2014/2013-wijk-en-buurtkaart-art.htm


The coverage area of top 21 cities looks like this.

From the CBS data we learn that the top 21 cities cover 5020400 out of 16828900 inhabitants. Which is29.83 %. In terms of area this is 2353 km2 of a total of 41545 km2, which is 5.66%.

For each city we will do the significance test separately in the next pages.


of 50


9.1 Gemeente Amsterdam

The table shows the speed test analysis for gemeente Amsterdam. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 4541 41852 2332 39789 0.011489 No 2062.4 +/- 2101.9 NAT-Mobile Vodafone 4541 41852 2437 39074 8.7035e-05 Yes 2777.9 +/- 1822.6 7.1


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 4541 18688 2332 15688 < 2.22e-16 Yes 3000.8 +/- 833.4 19.1T-Mobile Vodafone 4541 18688 2437 17735 0.0047396 Yes 953.6 +/- 869.7 5.4


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 4541 34.3 2332 41.8 1.1775e-07 Yes -7.4 +/- 3.6 -17.7T-Mobile Vodafone 4541 34.3 2437 42.1 3.2969e-08 Yes -7.8 +/- 3.6 -18.5


Please note the column [Sign.] or “Significant”, if this column states “No” the test is not significant atthe 99% confidence level. This means there is no significant difference in means observed so the NULLHypothesis can not be rejected. In layman’s terms: You cannot conclude that one operator has a faster4G network then the other on that particular 4G network speed measurement (download speed, uploadspeed or latency).

Explanation of termsSample 1: Number of speedtest samples for operator 1.

Sample 2: Number of speedtest samples for operator 2.

Mean 1: Average speed of speedtests for operator 1 in KBps. A high number here means that this operator has a fastdownload(or upload) speed. For the latency a high number means you have to wait longer to contact webpages or servers.






Rel(%): Relative difference in percentage. It is calculated as the difference of the mean divided by the slower of the twooperators average speed. If the difference is not significant (colums Sign is No), this column will state NA (not applicable).The comparison rules are similar to what is explained in the Diff(in Kbps or ms).


of 50


9.2 Gemeente Rotterdam

The table shows the speed test analysis for gemeente Rotterdam. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 1845 40350 1786 48123 7.2602e-14 Yes -7772.9 +/- 2665.6 -19.3T-Mobile Vodafone 1845 40350 1220 41619 0.24484 No -1268.7 +/- 2811.6 NA


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 1845 19639 1786 18549 0.012725 No 1090 +/- 1127 NAT-Mobile Vodafone 1845 19639 1220 17964 0.00095449 Yes 1675.1 +/- 1305.5 9.3


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 1845 33.3 1786 39.4 1.0665e-09 Yes -6.1 +/- 2.6 -15.5T-Mobile Vodafone 1845 33.3 1220 43.6 < 2.22e-16 Yes -10.3 +/- 3.1 -23.6













of 50


9.3 Gemeente ’s-Gravenhage

The table shows the speed test analysis for gemeente ’s-Gravenhage. For this test the confidence level isset to 99%.



Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 2189 20359 1630 17595 9.9484e-11 Yes 2763.6 +/- 1098.1 15.7T-Mobile Vodafone 2189 20359 965 14884 < 2.22e-16 Yes 5474.9 +/- 1282.8 36.8


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 2189 35.4 1630 40.9 0.00086262 Yes -5.5 +/- 4.2 -13.4T-Mobile Vodafone 2189 35.4 965 49.4 1.3342e-12 Yes -13.9 +/- 5 -28.1













of 50


9.4 Gemeente Utrecht

The table shows the speed test analysis for gemeente Utrecht. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 834 38923 1033 40794 0.18653 No -1870.4 +/- 3649.7 NAT-Mobile Vodafone 834 38923 849 35268 0.0053994 Yes 3655.8 +/- 3384 10.4


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 834 16853 1033 16926 0.89592 No -73 +/- 1439.2 NAT-Mobile Vodafone 834 16853 849 15888 0.10941 No 965.2 +/- 1554 NA


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 834 42.9 1033 41.3 0.72156 No 1.6 +/- 11.6 NAT-Mobile Vodafone 834 42.9 849 42.2 0.87213 No 0.7 +/- 11.5 NA













of 50


9.5 Gemeente Eindhoven

The table shows the speed test analysis for gemeente Eindhoven. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 647 45718 934 51300 0.00032797 Yes -5582.6 +/- 3999.2 -12.2T-Mobile Vodafone 647 45718 631 35104 4.0538e-12 Yes 10613.1 +/- 3908.7 30.2

















of 50


9.6 Gemeente Tilburg

The table shows the speed test analysis for gemeente Tilburg. For this test the confidence level is set to99%.



Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 468 17526 445 15673 0.023148 No 1852.5 +/- 2102 NAT-Mobile Vodafone 468 17526 352 14090 5.5872e-05 Yes 3435.8 +/- 2189.8 24.4


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 468 37.2 445 43.5 0.0086167 Yes -6.2 +/- 6.1 -14.3T-Mobile Vodafone 468 37.2 352 44 0.0028955 Yes -6.8 +/- 5.9 -15.5













of 50


9.7 Gemeente Groningen

The table shows the speed test analysis for gemeente Groningen. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 384 45344 831 49124 0.047627 No -3780.2 +/- 4919.6 NAT-Mobile Vodafone 384 45344 456 28684 < 2.22e-16 Yes 16659.7 +/- 4966.8 58.1


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 384 17647 831 18643 0.21723 No -996.3 +/- 2083.7 NAT-Mobile Vodafone 384 17647 456 14093 2.4844e-05 Yes 3554.4 +/- 2162.9 25.2


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 384 45.3 831 39.2 0.29653 No 6.2 +/- 15.3 NAT-Mobile Vodafone 384 45.3 456 48.7 0.57308 No -3.4 +/- 15.5 NA













of 50


9.8 Gemeente Almere

The table shows the speed test analysis for gemeente Almere. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 400 42471 547 39494 0.13018 No 2976.9 +/- 5072.7 NAT-Mobile Vodafone 400 42471 331 23145 < 2.22e-16 Yes 19325.6 +/- 4147.3 83.5




Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 400 33.8 547 40.3 0.029834 No -6.5 +/- 7.8 NAT-Mobile Vodafone 400 33.8 331 45.3 0.00022966 Yes -11.5 +/- 8 -25.4













of 50


9.9 Gemeente Breda

The table shows the speed test analysis for gemeente Breda. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 373 38843 415 31444 8.6577e-06 Yes 7399 +/- 4266.1 23.5T-Mobile Vodafone 373 38843 294 29768 1.7866e-06 Yes 9075.5 +/- 4862.6 30.5


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 373 16909 415 13415 1.9089e-05 Yes 3493.5 +/- 2095.4 26T-Mobile Vodafone 373 16909 294 14384 0.0067426 Yes 2524.8 +/- 2399.8 17.6


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 373 51.5 415 44 0.52871 No 7.5 +/- 30.7 NAT-Mobile Vodafone 373 51.5 294 46 0.64635 No 5.4 +/- 30.6 NA













of 50


9.10 Gemeente Nijmegen

The table shows the speed test analysis for gemeente Nijmegen. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 309 41077 599 35553 0.007439 Yes 5523.8 +/- 5315 15.5T-Mobile Vodafone 309 41077 305 28694 9.16e-08 Yes 12382.5 +/- 5916.3 43.2


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 309 17491 599 15827 0.074565 No 1663.8 +/- 2407.2 NAT-Mobile Vodafone 309 17491 305 13805 0.00024274 Yes 3686.2 +/- 2579.7 26.7


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 309 38.6 599 43.7 0.43573 No -5 +/- 16.7 NAT-Mobile Vodafone 309 38.6 305 43.5 0.45117 No -4.9 +/- 16.7 NA













of 50


9.11 Gemeente Enschede

The table shows the speed test analysis for gemeente Enschede. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 330 38398 504 45017 0.001861 Yes -6618.7 +/- 5474.1 -17.2T-Mobile Vodafone 330 38398 224 23802 < 2.22e-16 Yes 14596.6 +/- 4419.3 61.3


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 330 16536 504 17304 0.37359 No -768.6 +/- 2230 NAT-Mobile Vodafone 330 16536 224 13427 0.0013503 Yes 3108.4 +/- 2493.9 23.1


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 330 37.3 504 42.1 0.07578 No -4.8 +/- 7 NAT-Mobile Vodafone 330 37.3 224 50.6 0.10734 No -13.2 +/- 21.2 NA













of 50


9.12 Gemeente Apeldoorn

The table shows the speed test analysis for gemeente Apeldoorn. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 284 46905 673 38032 2.231e-05 Yes 8873.1 +/- 5367.8 23.3T-Mobile Vodafone 284 46905 279 31712 7.2824e-11 Yes 15192.7 +/- 5910.9 47.9


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 284 18407 673 14712 6.946e-05 Yes 3695.3 +/- 2379.8 25.1T-Mobile Vodafone 284 18407 279 13055 3.1758e-07 Yes 5351.7 +/- 2671.2 41


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 284 33.9 673 44.8 0.00018344 Yes -10.9 +/- 7.5 -24.3T-Mobile Vodafone 284 33.9 279 45.8 0.00081078 Yes -11.8 +/- 9.1 -25.8













of 50


9.13 Gemeente Haarlem

The table shows the speed test analysis for gemeente Haarlem. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 399 44050 310 36116 0.00041186 Yes 7934.2 +/- 5771.1 22T-Mobile Vodafone 399 44050 312 26661 < 2.22e-16 Yes 17388.9 +/- 5037.5 65.2


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 399 20863 310 13732 2.3733e-13 Yes 7131.1 +/- 2465.3 51.9T-Mobile Vodafone 399 20863 312 11595 < 2.22e-16 Yes 9267.9 +/- 2442.3 79.9















of 50


9.14 Gemeente Amersfoort

The table shows the speed test analysis for gemeente Amersfoort. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 285 38894 468 24258 6.2876e-15 Yes 14635.8 +/- 4717.5 60.3T-Mobile Vodafone 285 38894 295 34740 0.06598 No 4154 +/- 5828.2 NA


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 285 14180 468 14967 0.30887 No -786.5 +/- 1995.9 NAT-Mobile Vodafone 285 14180 295 14827 0.46598 No -646.7 +/- 2290.9 NA


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 285 41.5 468 40.5 0.74477 No 1 +/- 8.2 NAT-Mobile Vodafone 285 41.5 295 41.9 0.90489 No -0.4 +/- 8.7 NA













of 50


9.15 Gemeente Arnhem

The table shows the speed test analysis for gemeente Arnhem. For this test the confidence level is set to99%.



Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 513 19118 539 16320 0.00021898 Yes 2797.5 +/- 1945.9 17.1T-Mobile Vodafone 513 19118 309 12597 1.5539e-14 Yes 6520.8 +/- 2150.3 51.8


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 513 37.7 539 39.1 0.79539 No -1.4 +/- 14.4 NAT-Mobile Vodafone 513 37.7 309 41.3 0.51825 No -3.6 +/- 14.4 NA













of 50


9.16 Gemeente Zaanstad

The table shows the speed test analysis for gemeente Zaanstad. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 300 46343 389 50239 0.13546 No -3896.3 +/- 6733.3 NAT-Mobile Vodafone 300 46343 228 29863 6.6098e-11 Yes 16479.4 +/- 6385.9 55.2


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 300 18966 389 17121 0.068476 No 1845 +/- 2612.1 NAT-Mobile Vodafone 300 18966 228 14625 5.0837e-05 Yes 4340.6 +/- 2746.6 29.7


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 300 33.1 389 39.6 0.011674 No -6.5 +/- 6.6 NAT-Mobile Vodafone 300 33.1 228 39 0.015463 No -5.9 +/- 6.3 NA













of 50


9.17 Gemeente Haarlemmermeer

The table shows the speed test analysis for gemeente Haarlemmermeer. For this test the confidence levelis set to 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 643 37819 668 33435 0.0019843 Yes 4384.3 +/- 3649.6 13.1T-Mobile Vodafone 643 37819 507 33038 0.0028777 Yes 4781 +/- 4129.4 14.5


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 643 18052 668 15048 1.3972e-05 Yes 3003.9 +/- 1776.6 20T-Mobile Vodafone 643 18052 507 14915 1.6399e-05 Yes 3137.4 +/- 1870.6 21


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 643 35.9 668 42.8 0.014813 No -6.9 +/- 7.3 NAT-Mobile Vodafone 643 35.9 507 48 0.00018753 Yes -12.1 +/- 8.3 -25.2













of 50


9.18 Gemeente ’s-Hertogenbosch

The table shows the speed test analysis for gemeente ’s-Hertogenbosch. For this test the confidence levelis set to 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 341 41447 428 40895 0.80116 No 552.3 +/- 5661.6 NAT-Mobile Vodafone 341 41447 333 34930 0.0037256 Yes 6517.7 +/- 5784.2 18.7


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 341 16088 428 16040 0.95403 No 48.4 +/- 2166.7 NAT-Mobile Vodafone 341 16088 333 12874 0.00020028 Yes 3214.3 +/- 2219.7 25















of 50


9.19 Gemeente Zoetermeer

The table shows the speed test analysis for gemeente Zoetermeer. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 391 40092 401 34781 0.0034505 Yes 5311.7 +/- 4675.6 15.3T-Mobile Vodafone 391 40092 249 32211 0.00040427 Yes 7881.4 +/- 5720 24.5


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 391 18267 401 15261 0.00026166 Yes 3005.8 +/- 2116 19.7T-Mobile Vodafone 391 18267 249 12561 6.5808e-09 Yes 5705.8 +/- 2504.2 45.4


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 391 33 401 40.5 0.0024517 Yes -7.5 +/- 6.4 -18.5T-Mobile Vodafone 391 33 249 51.5 6.3947e-09 Yes -18.5 +/- 8.1 -35.9













of 50


9.20 Gemeente Zwolle

The table shows the speed test analysis for gemeente Zwolle. For this test the confidence level is set to99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 262 43224 424 30162 4.1998e-09 Yes 13062.7 +/- 5642.6 43.3T-Mobile Vodafone 262 43224 274 27236 2.6972e-13 Yes 15988.2 +/- 5479.7 58.7


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 262 16257 424 13789 0.0089638 Yes 2467.6 +/- 2431.8 17.9T-Mobile Vodafone 262 16257 274 14410 0.070225 No 1846.4 +/- 2631.3 NA


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 262 34.7 424 41 7.0096e-08 Yes -6.3 +/- 3 -15.4T-Mobile Vodafone 262 34.7 274 45 1.849e-11 Yes -10.3 +/- 3.9 -22.9













of 50


9.21 Gemeente Maastricht

The table shows the speed test analysis for gemeente Maastricht. For this test the confidence level is setto 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 219 40525 214 35440 0.047859 No 5084.6 +/- 6629.8 NAT-Mobile Vodafone 219 40525 577 31166 9.3139e-06 Yes 9358.6 +/- 5394 30


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 219 17492 214 16354 0.31502 No 1138.2 +/- 2927.5 NAT-Mobile Vodafone 219 17492 577 15745 0.064254 No 1747.5 +/- 2438.3 NA


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 219 39.1 214 48 1.7453e-14 Yes -8.9 +/- 2.9 -18.5T-Mobile Vodafone 219 39.1 577 61.8 3.7937e-08 Yes -22.7 +/- 10.5 -36.7













of 50


9.22 Gemeente Unkown Location

The table shows the speed test analysis for gemeente Unkown Location. For this test the confidence levelis set to 99%.

Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 7461 44715 14828 35141 < 2.22e-16 Yes 9573.2 +/- 1018.4 27.2T-Mobile Vodafone 7461 44715 9180 28259 < 2.22e-16 Yes 16456 +/- 1053.2 58.2


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(Kbps) Conf Int Rel(%)T-Mobile KPN 7461 17289 14828 14143 < 2.22e-16 Yes 3145.5 +/- 457.4 22.2T-Mobile Vodafone 7461 17289 9180 13967 < 2.22e-16 Yes 3322 +/- 502.6 23.8


Operator1 Operator2 Sample1 Mean1 Sample2 Mean2 P-value Sign. Diff(ms) Conf Int Rel(%)T-Mobile KPN 7461 36.3 14828 39.7 2.8795e-05 Yes -3.4 +/- 2.1 -8.6T-Mobile Vodafone 7461 36.3 9180 43.6 < 2.22e-16 Yes -7.4 +/- 2.1 -17













of 50

independent speed test analysis of 4g mobile networks …€¦ ·...

Documents